1. Field of the Invention
This invention relates to a single crystal gallium nitride (GaN) substrate, a method of growing single crystal GaN and a method of making a single crystal GaN substrate utilized as a substrate of making laser diodes (LDs) and light emitting diodes (LEDs) composed of groups 3–5 nitride semiconductors.
This application claims the priority of Japanese Patent Applications No.2001-311018 filed on Oct. 9, 2001 and No.2002-269387 filed on Sep. 17, 2002, which are incorporated herein by reference.
2. Description of Related Art
Light emitting devices based upon group 3–5 nitride semiconductor include blue/green light emitting diodes and blue light laser diodes. Blue light LEDs have been sold on the market. But, LDs have not been on the market yet. Almost all of the conventional 3–5 nitride light emitting devices and laser diodes (LEDs, LDs) have been fabricated upon sapphire (α-Al2O3) substrates. Sapphire is a rigid and sturdy material. Sapphire excels in chemical and physical stability. Another advantage of sapphire is to allow GaN heteroepitaxial growth on it. Thus, GaN films, AlGaN films or InGaN films can be grown on sapphire substrates. Sapphire has been an exclusive, pertinent substrate for GaN type LEDs.
Sapphire, however, has some drawbacks as a substrate. Sapphire lacks cleavage. Sapphire is not a semiconductor but an insulator. GaN films or InGaN films grown on a sapphire substrate are annoyed by large lattice misfitting. Lattice misfitting means a difference of lattice constants between a substrate and a film. Sapphire belongs to trigonal symmetry group. Sapphire lacks three-fold rotation symmetry and inversion symmetry. Poor symmetry deprives sapphire of cleavage planes.
The use of sapphire substrates forces device makers to cut a processed GaN wafer into individual chips by mechanical dicing instead of natural cleavage. To dice a hard, sturdy, rigid sapphire plate mechanically into pieces is a difficult process, which decreases yield and enhances cost.
Noncleavage further induces a serious difficulty of making good resonator mirrors of laser diodes. The resonators are made by mechanical polishing, which raises the cost of LDs and declines the quality of the resonators.
Insulation is another weak point of sapphire. Insulating sapphire incurs a difficulty of n-electrodes. An insulating substrate forbids an LED from having an n-electrode on the bottom unlike an ordinary diode. An n-electrode is formed by etching away a top p-GaN layer, an active layer, revealing an intermediate n-GaN film on the sapphire substrate, depositing an n-metal electrode on the n-GaN film, and wirebonding the n-metal electrode with a lead pin. The etching for revealing the intermediate film and wirebonding are extra steps which are required for making an n-electrode on the on-sapphire device.
Current flows in a horizontal direction in the n-GaN film. The n-GaN film should be grown to a thick film for reducing electric resistivity of the n-GaN film. Extra steps and extra components raise the cost of fabrication.
Since two electrodes are formed on n- and p-films within a chip, an extra wide area is required for the chip. The wide, large chip raises cost up.
The third weak point of a sapphire substrate is lattice misfitting. Lattice misfitting induces high density dislocations into GaN epi-layers grown upon a sapphire substrate. It is said that GaN epi-layers of on-sapphire LEDs sold on the market should have 1×109 cm−2 dislocations.
Another candidate for a substrate is silicon carbide SiC, since lattice misfit between SiC and GaN is smaller than the GaN/sapphire misfit. A GaN grown on a SiC substrate turns out to have a similar high dislocation density to the on-sapphire GaN layers. SiC does not surpass sapphire as a substrate.
High dislocation density in GaN, InGaN, AlGaN epi-layers causes no problem in the nitride-type LEDs because of low current density. In the case of LDs having a narrow striped electrode and a narrow emission area, high density current would reproduce dislocations and the increased dislocations would shorten the lifetime of LDs. Non-cleavage, insulator and misfit are three serious drawbacks of sapphire substrates.
The best candidate for an ideal substrate for LDs is a gallium nitride (GaN) single crystal substrate. If a high quality GaN single crystal substrate were obtained, the problem of the lattice misfitting would be solved, because a device would take a GaN/GaN homoepitaxy structure.
A GaN crystal has cleavage planes {1−100}. Cleavability of GaN enables device makers to divide a processed GaN wafer into individual chips along cleavage planes. Cleavage lowers the difficulty and cost of chip separation. Resonator mirrors of LDs can be easily produced by natural cleavage. High quality resonators are formed by the cleavage.
GaN can be doped with n-type dopants or p-type dopants. Doping with an impurity can prepare a conductive GaN substrate. Since a low resistance n-type substrate is made by doping with an n-type dopant, an n-electrode can be formed at a bottom of an n-GaN substrate. Vertical electrode alignment enables an LD and an LED to reduce a chip size, simplify a device structure and curtail the cost.
However, GaN single crystals are not yielded as natural resources. Production of GaN single crystals is difficult. Manufacture of high quality GaN single crystal substrates with a practical size has been impossible till now.
It is said that ultrahigh pressure and ultrahigh temperature would realize production of a tiny GaN crystal grown from a mixture of melt/solid at thermal equilibrium. The ultrahigh pressure method is impractical. A wide GaN substrate cannot be made by the method.
Methods of making GaN substrate crystals by growing a thick GaN crystal on a foreign material substrate in vapor phase and eliminating the foreign material substrate had been proposed. The vapor phase method has been inherently a method for making thin GaN, AlGaN, InGaN films on a sapphire substrate. The vapor phase method was diverted from film piling to substrate production. The inherent vapor phase method is unsuitable for substrate production. Large inner stress and many dislocations appeared in the GaN films made by the vapor phase method. Large inner stress prevented GaN films from growing thick crystals sufficient for substrates. A GaN “substrate” is a final product of the present invention. A substrate of a foreign material, e.g., sapphire or GaAs is a starting base plate for making GaN. Two substrates should not be confused. For discriminating two kinds of substrates, the starting foreign substrate is here named “undersubstrate”.
The inventors of the present invention proposed an epitaxial lateral overgrowth method of growing a GaN via a mask on an undersubstrate in vapor phase ({circle around (1)}Japanese Patent Application No.9-298300, {circle around (2)}Japanese patent application No.10-9008).
In the concrete, the ELO method proposed by us was a method of preparing a GaAs undersubstrate, producing an SiO2 or SiN film on the GaAs undersubstrate, perforating many small windows regularly and periodically aligning with a short pitch (spatial period), growing a GaN film on the masked GaAs substrate in vapor phase for a long time, and eliminating the GaAs substrate. The ELO alleviates inner stress and dislocations. The preceding ELO method utilized sapphire as an undersubstrate, which may be called an on-sapphire ELO. But, the above ELO method made use of GaAs as an undersubstrate. The method of the present inventors is called here an on-GaAs ELO.
The inventors of the present invention have proposed a method of making a plurality of GaN substrates by homoepitaxially growing a thick GaN crystal upon a GaN substrate obtained by the former mentioned ELO method, making a tall GaN ingot and slicing the tall GaN ingot into a plurality of wafers ({circle around (3)}Japanese Patent Application No.10-102546).
The improved ELO gave a probability for making wide GaN single crystal substrates on a commercial scale.
The ELO made GaN crystals were plagued with high dislocation density. The ELO reduces dislocations at an early stage of the growth. During the long time growth, however, dislocations increase again. Bad quality prohibited the ELO-GaN substrate from being the substrates for producing nitride type laser diodes (InGaN-LDs). Production of high quality (long lifetime) LDs required lower dislocation density GaN substrates.
Mass production of devices requires wide GaN substrates which have low dislocation density and high quality in a wide area.
The inventors of the present invention proposed a method of making low dislocation density GaN substrate ({circle around (4)}Japanese Patent Laying Open No.2001-102307). The present invention is an improvement of the former method {circle around (4)}.
The method proposed by {circle around (4)} is now called “facet growth” method in short. The method reduces dislocations by forming three-dimensional facets and facet pits of e.g., reverse-hexagonal cones on a growing surface intentionally, maintaining the facets and pits, growing a GaN crystal without burying the pits, gathering dislocations by the facets to the bottom of the pits, and reducing dislocations in other regions except the pit bottoms.
Three-dimensional facet pits are otherwise reverse-dodecagonal cones built by facets. The facets comprise typical {11−22} and {1−101} planes.
The facet growth {circle around (4)} (Japanese Patent Laying Open No.2001-102307) proposed by the inventors grows a GaN crystal in vapor phase on the condition of making facets and maintains the facets without burying the pits of facets. The facets grow not in the c-axis direction but in a direction normal to the facets. The roles of facets and pits in the facet growth {circle around (4)} (Japanese Patent Laying Open No.2001-102307) are described with reference to FIG. 1 which shows a small part around a facet pit on a surface of a GaN crystal growing in the facet growth. In practice, many facets and facet pits appear on the surface. A vapor phase epitaxy method (HVPE, MOCVD, MOC or Sublimation) grows a GaN crystal on a substrate in a direction of a c-axis. The growth is a c-axis direction growth but is not a “C-plane growth” which has been prevalent in the conventional GaN growth. Facets grow in directions normal to the facets.
Conventional C-plane growth methods grow a GaN film epitaxially on a substrate by maintaining a smooth C-plane surface. Produced GaN crystals have poor quality of high dislocation density, for example, 1010 cm−2. Our new facet growth method intentionally makes facets and pits, maintains the facets and reduces dislocations by make the best use of the function of facets of gathering dislocations into pit bottoms.
The facet growth produces plenty of reverse hexagonal cone pits 4 on the growing GaN surface. FIG. 1 shows a single one of many pits. Six slanting planes are low index facets 6 of {11−22} or {1−101} planes. A flat top surface 7 outside of the pit 4 is a surface of C-plane growth. In the pit, the facet grows inward in the direction of a normal standing on the facet as shown by inward arrows. Dislocations are swept to corner lines 8 by the growing facet. Dislocations are gathered on the six corner lines 8.
The dislocations swept to the corner lines 8 slide down along the corner lines to the bottom of the pit. In practice, the dislocations do not fall along the corner lines 8. The growth raised the facets, the corner lines and the pit bottoms at a definite speed. Sliding dislocations along the rising corner lines centrifugally move inward in horizontal directions. Finally, the dislocations attain to the center of the pit just at the time when the pit bottom rises to the height of the dislocation. Then, dislocations are accumulated at the bottom of the pit. The number of the dislocations on the facets is reduced by the accumulation of dislocations at the bottom.
Proceeding of the facet growth sometimes forms planar defects 10 following the corner lines 8 by storing the swept dislocations at the corner lines. The planar defects are six planes with sixty degree rotation invariance corresponding with the hexagonal symmetry of GaN. The width of the planar defects 10 is equal to the diameter of the pit 4. The six planar defects 10 cross at a vertical extension of the pit bottom. The crossing line forms a linear defect assembly 11 having highly concentrated dislocation. Ideally all the dislocations initially existing in the pit are swept to and are accumulated at the pit bottom. The other parts lose dislocations and become low dislocation density single crystals. This is the dislocation reduction method proposed by {circle around (4)}(Japanese Patent Laying Open No.2001-102307).
Finally, the majority of dislocations are concentrated to the pit center. The operations of the facets reduce dislocation density in the regions included within the projection of the pits.
There are some problems in the new facet growth method proposed by {circle around (4)} which makes facet pits at random spots accidentally, maintains the facet pits, grows a GaN crystal without burying the pits, and concentrates dislocations to the bottoms of the facet pits.
Though the facets gather dislocations to the pit bottoms, dislocations are not concentrated fully into a narrow, restricted spot. For example, when 100 μmφ pits are yielded, some pits can concentrate dislocations to a small spot at the bottom of a several micrometer diameter but other pits have about 50 μmφ hazy dislocation dispersion region of medium dislocation density near the bottom.
FIG. 3 demonstrates the occurrence of the hazy dislocation dispersion. FIG. 3(1) shows that a c-axis crystal growth (arrows) moves facets 16 inward, dislocations on the facets are carried by the facets 16 in horizontal directions (shown by horizontal lines) to the pit bottom and the bottom has a linear dislocation bundle 15. But, repulsive forces release once gathered dislocations outward. FIG. 3(2) shows that the once concentrated dislocations 15 are diffusing from the bottom to the facet 16 of a pit 14. Occurrence of hazy dislocation dispersion is a drawback of the facet growth of {circle around (4)}.
If the pit size is enlarged for widening the area of good quality portions, the area of the hazy dislocation dispersion further dilates. The reason is supposed that enlargement of a pit size increases the number of the dislocations gathered at the bottom and the number of the dislocations released from the bundle.
The inventors think that the release of dislocations from the pit centers results from repulsion acting between concentrated dislocations. Unification of pits incurs disorder of dislocations and expansion of the hazy dispersion of dislocations. Excess concentration induces the hazy dislocation dispersion.
The hazy dislocation dispersion has about 2×107 cm−2 dislocation density which has dependence to positions. Such a high dislocation density GaN substrate is insufficient for making laser diodes (LDs) of a satisfactory lifetime. A long lifetime of LDs requires to reduce dislocations down to one twentieth ( 1/20) of the current value (2×107 cm−2), that is, to 1×106 cm−2.
Another problem is the existence of planar defects 10 produced under the corner lines of pits as shown in FIG. 1(b). The planar defects are radially arranged with 60 degree rotation symmetry. Facets assemble dislocations at pit corner lines. Without progressing to the center bottom, the assembled dislocations form planar defects 10 by dangling from the corner lines. A planar defect can be considered as an alignment of dislocations in a plane. The planar defects are another problem of the conventional facet growth method. Sometimes a slide of crystal planes occurs on both sides of the planar defect.
Besides the 60 degree rotation symmetric planar defects, 30 degree rotation symmetry planar defects sometimes appear in dodecagonal pits on a growing surface. Planar defects appear as dislocation arrays on the surface of the growing substrate. Planar defects are a serious hindrance to produce long lifetime LDs. Prolongation of LD lifetime requires reduction of the planar defects.
The final problem is distribution of defects. Dislocation reduction of the facet growth method makes use of facet pits accidentally and randomly appearing on a facet growth. Positions of pits are not predetermined. Numbers of appearing pits are also not programmable. Positions, numbers, shapes and sizes of appearing pits are all stochastic, random, accidental variables which are unpredeterminable, unprogrammable, uncontrollable. It is a problem that the positions of pits are uncontrollable.
If a plurality of laser diodes were fabricated upon a GaN substrate having random distributing planar defects, emission stripes of active layers of the laser diodes would accidentally coincide with the defect assemblies which occupy random spots on the GaN substrate. In the case of coincidence of the active layer with the defect bundles, important emission layers are plagued by the defect assemblies. Large current density driving current would invite rapid degeneration on emission stripes from the inherent defects of the laser diodes.
Uncontrollability of the positions of pits would decrease the yield of manufacturing laser diodes on the substrate.
Manufacturing GaN substrates for making laser diodes thereon requires enhancement of yield through controlling the positions of dislocation bundles on the GaN substrates. It is important to control the positions of dislocation bundles not to collide with emission stripes of laser diode chips on the GaN substrates.
Three problems have been described for long lifetime laser diodes. The purpose of the present invention is to conquer the three problems;    (1) Reduction of hazy dislocation diffusion from the pit center composed of facets,    (2) Extinction of planar defects at the bottoms of the pits composed of facets,    (3) Controlling of positions of the pits made of facets.
The present invention aims at overcoming the difficulties of the three problems. Preliminary descriptions are given to orientations of crystals and vapor phase growth of gallium nitride (GaN) for facilitating the understanding of the present invention. The present invention can be carried out by any of the vapor phase methods described here. GaN has hexagonal symmetry. Designation of planes and directions of GaN is far more difficult than cubic symmetry, e.g., silicon (Si) or gallium arsenide (GaAs).
Clear understanding of the definitions of crystal planes, directions, and orientations is indispensable for describing relations of parts and structures of GaN crystals. Three index designation and four index designation are employed for expressing planes and directions of GaN crystal. Here, the four index designation is chosen.
There are some rules for determining expressions of crystal planes and crystal directions. Integers h, k, m and n are used for representing planes. The integers are called “Miller indices” or plane indices. Collective representation of planes is taken into wavy brackets as {hkmn} without comma. Individual representation of planes is taken into round brackets as (hkmn) without comma. Collective representation of directions is taken into key brackets as <hkmn> without comma. Individual representation of directions is taken into square brackets as [hkmn] without comma. The four kinds of brackets for representation should be clearly discriminated. An individual plane (hkmn) is orthogonal to an individual direction having the same Miller indices [khmn]. Namely, a [hkmn] direction is a normal of a (hkmn) plane.
Allowable symmetry operations are determined by the symmetry group of the object crystal. When a plane or direction is returned to another plane or direction by an allowable symmetry operation, two planes or two directions are equivalent. Equivalent planes or directions are represented by a common collective representation. Hexagonal GaN has three time rotation symmetry, which allows commutation of three indices h, k, m of a-, b- and d-axes. Miller indices h, k and m are equivalent. The final index n of a c-axis is a unique one which cannot be commutated with other indices. A collective plane representation {hkmn} includes all planes obtained by replacing a (hkmn) plane on all allowable symmetric operations.
Hexagonal symmetry group contains several different subgroups. Equivalent planes or directions depend upon the subgroup. GaN has three-fold rotation symmetry but lacks inversion symmetry. Sapphire (Al2O3) belongs not to hexagonal symmetry but to trigonal symmetry. Sapphire has neither three-fold rotation symmetry nor inversion symmetry. The following descriptions are valid only for GaN but invalid for sapphire without three-fold rotation.
GaN has three-fold rotation symmetry. Then, (hkmn), (kmhn), (mhkn), (hmkn), (khmn) and (mkhn) are six equivalent planes included in a collective representation {hkmn}. Six collective representations {hkmn}, {kmhn}, {mhkn}, {hmkn}, {khmn} and {mkhn} designate all the same planes. Miller indices are plus or minus integers. A minus sign should be denoted by an upper line. Since upper lines are forbidden in a patent description, a front “−” denotes a minus integer.
Since GaN has non inversion symmetry, {hkmn} is not identical to {−h−k−m−n}. A C-plane (0001) is different from a −C-plane (000−1) in GaN. Ga atoms exclusively appear on a C-plane (0001). But, N atoms exclusively appear on a −C-plane. Thus, a (0001) plane is sometimes denoted by a (0001) Ga plane and a (000−1) plane is sometimes denoted by a (000−1) N plane. The latter is often written as (0001) N plane by omitting a minus sign.
Hexagonal GaN has three equivalent principal axes having three-fold rotation symmetry. Two of the three axes are denoted by a-axis and b-axis. The third axis has no name traditionally. For the convenience of expression, the third axes is now called d-axis. Namely, the a-axis, b-axis and d-axis are defined with 120 degree angle rotation on a plane perpendicular to c-axis. The c-axis is a special axis different from the three axes in hexagonal symmetry. Crystal planes are an assembly of indefinite number of parallel planes having a common inclination and a common distance. Miller indices of a plane are defined as reciprocals of a length of a segment of an axis cut by a first plane divided by the axis length. When the first plane cuts a-axis at a/h, cuts b-axis at b/k, cuts d-axis at d/m and cuts c-axis at c/n, the set of the planes is denoted by (hkmn).
Smaller Miller index planes are more important planes with smaller number of equivalent planes. Smaller index planes appear on a crystal surface more frequently than larger index planes. Larger Miller index planes are less important with large number of equivalent planes. Forward three indices are not independent, since the three indices include only two freedom. Three indices represent two-dimensional directions. Three indices can be represented by two indices at the sacrifice of symmetry. The three indices h, k and m are linearly dependent. Three indices h, k and m always satisfy a sum rule h+k+m=0.
GaN has three typical planes. One important plane is C-plane. C-plane is expressed by (0001) plane. C-plane is a plane which is perpendicular to c-axis. A plane (hkmn) is vertical to a direction [hkmn] having the same Miller indices. From now, planes are denoted by capital letters (C-, A-, M-planes) but directions are denoted by small letters (c-axis, a-axis, b-axis, d-axis) for discriminating planes from directions.
GaN which belongs to hexagonal symmetry has three-fold rotation symmetry which retrieves itself by 120 degree rotations around c-axis. C-plane (0001) has the highest symmetry. In the case of heteroepitaxy of GaN on a foreign material undersubstrate, a three-fold rotation symmetric plane of the foreign material should be utilized. GaN is grown on the undersubstrate in a c-axis direction for harmonizing the symmetry. GaN lacks inversion symmetry. (0001) plane and (000−1) plane are different plane. The discrimination between (0001) plane and (000−1) plane is later described.
The second important plane is called an M-plane which is a cleavage plane. An M-plane is a plane which crosses an edge of one of the three symmetric axes a, b and d is parallel to one of two other symmetric axes and is parallel to a c-axis. M-planes are represented by collective expressions of {1−100}, {01−10}, {−1010}, {−1100}, {0−110} and {10 −10 } or represented by individual expressions of (1−100), (01−10), (−1010), (−1100), (0−110) and (10−10).
Collective expressions are all equivalent. But, individual expressions signify different individual planes. Individual M-planes cross with each other at 60 degrees. It should be noted that not 90 degrees but 60 degrees are a crossing angle between individual planes. The M-plane is a convenient expression of an important plane of GaN.
The third important plane is called an A-plane. An A-plane is a plane which crosses two edges of two of the three symmetric axes a, b and d, and is parallel to a c-axis. A-planes are represented by collective expressions of {2−1−10 }, {−12−10}, {−1−120 }, {−2110}, {1−210} and {11−20} or represented by individual expressions of (2−1−10), (−12−10), (−1−120), (−2110), (1−210) and (11−20).
GaN lacks six-fold rotation symmetry. The above six individual planes signify two kinds of planes. The A-plane is a convenient expression of denoting the important plane. The A-plane should not be confused with the a-axis. The A-planes are not rectangular to the a-axis.
A direction <2−1−10> which has the same Miller indices as an A-plane (2−1−10) is perpendicular to the A-plane. But, the direction <2−1−10> is not called an a-direction. A direction <1−100> which is perpendicular to an M-plane (1−100) is not called an m-direction.
A GaN crystal has three typical, important planes; C-plane, A-plane and M-plane. Don't confuse directions with planes. A direction and a plane with the same Miller indices are perpendicular. On the contrary, a direction and a plane with vertical Miller indices (hh′+kk′+mm′=0) are parallel.
“Facet” is an important word which appears frequently in this specification. Facets inherently mean small planes appearing on a finished crystal. In the description, facets are planes obtained by slightly slanting A-planes or M-planes toward the c-axis. Sometimes the facets which are attained by slanting A-planes are called “A-derivative” facets. {11−21} and {11−22} are A-derivative facets. The facets which are attained by slanting M-planes are called “M-derivative” facets. {1−10} and {1−102} are M-derivative facets.
A V-groove (valley) is formed by two crossing planes having common forward three indices h, k, m and different fourth index n. Typical valleys are made, for example, by A-derivative facets {2−1−1±1} or {2−1−1±2}. Other typical valleys are formed by M-derivative facets {1−10±1} or {1−10±2}.
The fourth index is either 2 or 1 in the above examples of the V-grooves. Lower fourth index planes appear with higher probability. The fourth index n means an inclination to the c-axis. {2−1−11} facets are obtained by slanting {2−1−10} A-planes slightly to the c-axis. {2−1−12} facets are obtained by further slanting {2−1−11} facets to the c-axis. Higher four index n means a larger slanting angle to the c-axis and a smaller inclination to the horizontal plane; C-plane (0001). Probable values of the forth index n are n=1, 2, 3 and 4.
In many cases, a V-groove is formed by one-step facets. A concept of two-step facets will appear later. A V-groove is sometimes formed by two different slope facets. Upper pairs of facets are bigger and steeper, which has a smaller n. The upper facets are called “groove-facet”. Lower pair of facets have smaller and milder, which has a larger n. The lower facets of a V-groove are called “shallower” facets.
V-grooves (valleys) are formed mainly by M-derivative {11−22} facets or A-derivative {1−101} facets (groove facets, upper facets) in the present invention. The length of the a-axis (=b-, d-axis) is denoted by “a”. The length of the c-axis is denoted by “c”. An inclination of {1−101} facets to the C-plane is given by tan−1(31/2a/2c). An inclination of {11−22} facets to the C-plane is given by tan−1(a/c).
Shallower facets appearing at bottoms of pits are denote by {11−23}, {1−102}, {11−24}, {1−103} which have index n of higher values. Slanting angles of {1−10n} planes (n≧2) to C-plane are tan−1(31/2 a/2cn). The slanting angles of tan−1(31/2a/2cn) for n larger than 2 are smaller than the slanting angle for n=1. Slanting angles of {11−2n} planes (n≧3) to C-plane are tan−1(2a/cn). The slanting angles of tan−1(2a/cn) for n larger than 3 are smaller than the slanting angle for n=2. The facet having a larger index n is denoted by a shallower facet.
GaN is a wurtzite type crystal of hexagonal symmetry. A GaN unit cell is a hexagonal column having a hexagonal bottom including seven Ga atoms positioned at six corners and a center point, a ⅜ height hexagonal plane including seven N atoms positioned at six corners and a center point, a ½ height hexagonal plane including three Ga atoms at corners of an equilateral triangle, a ⅞ height plane including three N atoms at corners of an equilateral triangle, and a hexagonal top including seven Ga atoms positioned at six corners and a center point. GaN has three-fold rotation symmetry. GaN, however, has neither inversion nor six-fold rotation symmetry.
Sapphire, silicon (Si), gallium arsenide (GaAs) wafers are used as an undersubstrate. Sapphire (α-Al2O3), trigonal symmetry, lacks three-fold rotation symmetry and inversion symmetry. Poor symmetry deprives sapphire of cleavage. Uncleavability is a serious drawback of sapphire.
Silicon has diamond-type cubic symmetry. Si has three Miller indices. Cubic symmetry enables three Miller indices k, h, m to define plane orientations (khm). Three Miller indices are independent. Unlike hexagonal symmetry, there is no sum rule among three Miller indices. Namely in general k+h+m≠0. Cubic symmetry has only one three-fold rotation symmetric direction. The direction is an orthogonal line direction, which is normal to a (111) plane. Usual silicon devices have been manufactured upon (001) plane substrates. But, the (001) plane lacks three-fold rotation symmetry. The (001) Si wafer cannot be a candidate for an undersubstrate of growing hexagonal GaN. Three-fold rotation symmetric Si (111) wafers can be a candidate for the undersubstrate.
GaAs is not hexagonal but cubic. GaAs has a zinc blende type (ZnS) lattice structure. Cubic GaAs is fully defined by three plane indices. GaAs has a unique three-fold rotation symmetry direction which is parallel to an orthogonal line. The three-fold symmetric plane is denoted by a (111) plane. Usual GaAs devices have been produced upon (001) planes which have cleavage planes (±1±10) perpendicular to the surface on four sides. But, the (001) wafer which lacks three-fold rotation symmetry cannot be an undersubstrate. Instead of (001), a (111) GaAs wafer can be a candidate for an undersubstrate for GaN growth.
GaAs lacks inversion symmetry. A (111) plane has two versions. One is a (111) plane having dangling As atoms. The other is a (111) plane having dangling g Ga atoms. The former is sometimes denoted by (111) As plane and the latter is denoted by (111) Ga plane. (111) Ga is otherwise represented by (111) A. (111) As is represented by (111) B.
The present invention employs a vapor phase growth for making GaN, for example, an HVPE method, an MOCVD method, an MOC method and a sublimation method. The methods are described.
[1. HVPE Method (Hydride Vapor Phase Epitaxy)]
HVPE employs metal gallium (Ga) as a gallium source unlike MOCVD or MOC. A nitrogen source is ammonia gas. The HVPE apparatus contains a vertical hot-wall furnace, a Ga-boat sustained at an upper spot in the furnace, a susceptor installed at a lower spot in the furnace, top gas inlets, a gas exhausting tube and a vacuum pump. An undersubstrate (sapphire etc.) is put on the susceptor. Metal Ga solids are supplied to the Ga-boat. The furnace is closed and is heated. The Ga solids are heated into a melt. Hydrogen gas (H2) and hydrochloric acid gas (HCl) are supplied to the Ga-melt. Gallium chloride (GaCl) is produced. Gaseous GaCl is carried by the hydrogen gas downward to the heated undersubstrate. Hydrogen gas (H2) and ammonia gas (NH3) are supplied to the gaseous GaCl above the susceptor. Gallium nitride (GaN) is synthesized and is piled upon the undersubstrate for making a GaN film. The HVPE has an advantage of immunity from carbon contamination, because the Ga source is metallic Ga and GaCl is once synthesized as an intermediate.
[2. MOCVD Method (Metallorganic Chemical Vapor Deposition)]
An MOCVD method is the most frequently utilized for growing GaN thin films on sapphire substrates. An MOCVD apparatus includes a cold wall furnace, a susceptor installed in the furnace, a heater contained in the susceptor, gas inlets, a gas exhaustion hole and a vacuum pump. A material for Ga is metallorganic compounds. Usually trimethyl gallium (TMG) or triethyl gallium (TEG) is employed as a Ga source. The material for nitrogen is ammonia gas. A substrate is placed upon the susceptor in the furnace. TMG gas, NH3 gas and H2 gas are supplied to the substrate on the heated susceptor. Reaction of ammonia and the TMG gas makes gallium nitride (GaN). GaN piles upon the substrate. A GaN film is grown on the substrate. This is the most prevalent way of making GaN films on sapphire substrates for producing InGaN-LEDs. The growing speed is low. If thick GaN crystals are made by the MOCVD, some problems occur. One is the low speed of growth. Another problem is low gas utility rate, which was not a problem for making thin films by consuming small amounts of material gases. The MOCVD requires excess amount of gas of ammonia. High rate of ammonia/TMG raises gas cost in the case of bulk crystal production due to a large consumption of gases. The low gas utility rate caused a serious problem in the case of making a thick GaN crystal. Another one is a problem of carbon contamination. The TMG (Ga-material) includes carbon atoms. The carbon atoms contaminate a growing GaN crystal. The carbon contamination degrades the grown GaN crystal, because carbon makes deep donors which lowers electric conductivity. The carbon contamination changes an inherently transparent GaN crystal to be yellowish.
[3. MOC Method (Metallorganic Chloride Method)]
A Ga material is a metallorganic material, for example, TMG (trimethylgallium) like the MOCVD. In the MOC, however, TMG does not react directly with ammonia. TMG reacts with HCl gas in a hot wall furnace. The reaction yields gallium chloride (GaCl) once. Gaseous GaCl is carried to a heated substrate. GaCl reacts with ammonia supplied to the substrate and GaN is synthesized and piled on the substrate. An advantage of this invention is small carbon contamination since GaCl is made at the beginning step. This method, however, cannot overcome the difficulty of excess gas consumption.
[4. Sublimation Method]
A sublimation method does not utilize gas materials but solid materials. The starting material is GaN polycrystals. The sublimation method makes a GaN thin film on an undersubstrate by placing polycrystalline GaN solid at a place and an undersubstrate at another place in a furnace, heating the furnace, yielding a temperature gradient in the furnace, subliming the solid GaN into GaN vapor, transferring the GaN vapor to the substrate at a lower temperature, and piling GaN on the substrate.
Before fundamental principles of the present invention are described, the three problems are clarified further.
A problem of the previous facet growth maintaining facet pits is a state of an assembly of dislocations. Propagation of dislocations on the facets in the pits sweeps and concentrates many dislocations to the center of the pit. The state of dislocation assemblies is unstable, which is a serious problem.
When two dislocations having different signs of Burgers vectors, which means a direction and a size of slipping of lattices, collide with each other, the dislocations sometimes vanish by occurrence of favorable cancellation. In practice, most of the dislocations swept by the same facet have Burgers vectors of same signs. No cancellation occurs between two dislocations of the same signs of Burgers vectors. Thus, the dislocations gathered to the dislocation assembly are scarcely cancelled by the reciprocal sign Burgers vectors. The converged dislocations do not vanish at the confluence of dislocations.
Repulsive force occurs between two dislocations of the same sign Burgers vectors. The repulsive force tends to release bundles of the once concentrated dislocations by giving the dislocations centrifugal forces. The dislocations diffuse outward by the repulsion. The diffusion yields hazy dispersion of dislocations in the vicinity of the dislocation bundles. The hazy dislocation dispersion is a problem.
The reason of making the hazy dislocation dispersion is no clear enough yet for the inventors. One reason is stress concentration due to the dislocation convergence. A plurality of pits are often coupled into a bigger pit during the growth. Coupling pits disturbs the arrangement of dislocations. Perturbation of the dislocation arrangement is another reason of the hazy dislocation diffusion occurring.
The number of assembled dislocations to the dislocation confluence increases. The increase of dislocations enlarges the hazy dislocation dispersion. Another reason is an increase of dislocations by the coupling of pits.
While dislocations gather to the center of the pits composed of facets, corner lines between neighboring facets yield six planar assemblies of dislocations hanging from the corner line, which lie along 6 radii which coincide with each other by 60 degree rotation. The planar defects hanging on the corner lines are generated by the facets sweeping dislocations to the six corner lines of hexagonal pits.
When the unification of pits enlarges a pit size, the number of the dislocations centripetally converging to the center increases, which enhances further the size of the planar defects. This is another drawback of the previous facet growth.
The positions of pit appearing are random, stochastic and accidental matters. Pits appear at random spots by chance. The positions of the facet pits are uncontrollable, stochastic and random.
When optoelectronic devices are produced upon a GaN substrate with the wide hazy dislocation dispersion, random dislocation assemblies fluctuate qualities of the devices, which decreases the yield of the device production.
The problems of the present invention are described again. The facet growth grows a GaN crystal by maintaining facets, sweeping dislocations on the facets to a bottom confluence and storing the dislocations at a narrow confluence. A problem is the non-convergence of dislocations and dislocation dispersion from the confluence. The dislocation dispersion would be solved by giving effective dislocation annihilation/accumulation devices in the GaN crystal.
Instead of a random narrow confluence following a pit, this invention intentionally makes regularly aligning defect assemblies as a dislocation annihilation/accumulation place. The present invention prepares dislocation annihilation/accumulation places by giving defect assemblies ruled by making defect assemblies at designed spots in a growing crystal.
The previous facet growth transports and converges dislocations by maintaining facets leading slopes. The function of conveying facets is not restricted in pit-shaped facets. Slopes of facets are important for sweeping dislocations. Shapes of a set of facets are less important. The inventors hit upon an idea of employing a linear set of facet strips instead of isolated conical facet pits.
The present invention makes a rack-shaped faceted surface having a number of linear valleys and hills aligning in parallel at a definite pitch, which looks like a series of triangle columnar prisms lying side by side.
FIG. 4, which is a section of a V-groove composed of facets, briefly demonstrates a method of the present invention. The same section continues in the direction vertical to the figure in FIG. 4. FIG. 3 is a section of a conical pit of the previous facet growth method. FIG. 4 sections are slightly similar to FIG. 3 sections. But, the actual shapes are quite different. Don't confuse the linearly continual FIG. 4 sections with the isolated FIG. 3 sections. An undersubstrate (not shown in the figures) allows a GaN crystal 22 to grow with facets 26 in a facet growth mode. A pair of complementarily inclining facets 26 and 26 forms a V-groove 24. Following the bottoms (valleys) 29 of the V-grooves 24, voluminous defect accumulating regions (H) grow upward. Low dislocation single crystal regions (Z) grow under slopes of the facets 26. There are flat tops 27 outside of the facet grooves 24. The flat top 27 is a C-plane. C-plane growth regions (Y) grow under the flat C-planes 27. The valleys 29 lead voluminous defect accumulating regions (H).
A facet 26 leads a low dislocation single crystal region (Z). A flat top 27 leads a C-plane growth region (Y). The C-plane growth regions (Y), which are low dislocation density single crystals, have electric resistance higher than that of the low dislocation single crystal regions (Z). Growing facets 26 sweep dislocations of the low dislocation single crystal regions (Z) and the C-plane 27 growth regions (Y) inward and converge the dislocations into the voluminous defect accumulating regions (H). Almost all of the dislocations centripetally run in parallel to the C-plane toward the voluminous defect accumulating regions (H). A part of dislocations couple and extinguish. The rest of the dislocations are arrested and accumulated in the voluminous defect accumulating regions (H). A voluminous defect accumulating region (H) consists of an inner core (S) and an interface (K). The dislocation annihilation/accumulation place is either a sole interface (K) or a set of a interface (K) and a core (S). The interface (K) or the core (S) never allow once-captivated dislocations to escape therefrom.
Unlike a narrow defect assembly 15 as shown in FIG. 3(1), the present invention prepares wide voluminous defect accumulating regions (H) and storing dislocations by the voluminous defect accumulating regions (H) with a definite thickness. Wideness and voluminousness enable the voluminous defect accumulating regions (H) to accommodate far more dislocations than the lean defect assembly 15 of FIG. 3(1). One advantage of the present invention is the vast capacity of the voluminous defect accumulating regions (H).
Instead of polygonal pits, the present invention employ linear facets aligning as wide strips extending in a definite direction. Six radial corner lines, which accompany polygonal pits, do not occur on a surface composed of the linear facets. The linear facets would not make radial planar defects. The present invention can avoid the difficulty of occurrence of planar defects, which is a drawback of the previous facet growth, by adopting linear facets.
It is confirmed that linear facet slopes enable linear polycrystalline regions with grain boundaries (K) to occur at the bottoms of the facets and the grain boundaries (K) to act as a dislocation annihilation/accumulation place.
The dislocation annihilation/accumulation places allow the present invention to eliminate the hazy dislocation diffusion from the confluence. The dislocation annihilation/accumulation places clear stagnating dislocations away from the narrow confluence. The dislocation annihilation/accumulation places also kill radial planar defects 10 as shown in FIG. 1(b).
The polycrystalline regions are suitable for the annihilation/accumulation places. The polycrystal character allows the voluminous defect accumulating regions (H) to accommodate much many dislocations. The inventors found out that the effective dislocation annihilation/accumulation place is not restricted to the polycrystalline regions (H).
Besides polycrystalline voluminous defect accumulating regions (H), some sorts of single crystal regions are also effective as the dislocation annihilation/accumulation places. Available single crystal regions (H) are a single crystal having an orientation slanting to the surrounding single crystal regions, a single crystal having an interface composed of planar defects, and a single crystal having an interface built with small inclination grain boundaries. Surprisingly, another single crystal (H) having an inverse polarity, which means the direction of a c-axis, is also available for a dislocation annihilation/accumulation place. Polycrystalline and single crystal voluminous defect accumulating regions (H) have a large volume with a definite width h. The large volume ensures large capacity of storing dislocations.
The dislocation annihilation/accumulation regions have a definite width h instead of an indefinitely thin regions (=planes). The definite thickness and volume of the annihilation/accumulation regions (H) have advantages over conventional ELO methods. A conventional epitaxial lateral overgrowth method (ELO) utilizing a similar stripe structure forms small facets, gathers dislocations by the facets to bisecting planes between neighboring windows, and stores the dislocations at the bisecting planes which become planar defects. The planar defects made by the conventional ELO have neither a sufficient thickness nor an enough volume, since the thickness of the planes is indefinitely small. Excess concentration of dislocations enhances the repulsion among dislocations, releases the dislocations and allows the dislocations to diffuse outward.
On the contrary, the present invention can produce the voluminous defect accumulating region (H) having a sufficient, definite thickness. The definite thickness produces two interfaces on both sides. Dislocations attracted from a left side are arrested and stored on a left side interface K1. Dislocations attracted from a right side are gathered and accommodated on a right side interface K2. Dislocations are divided into halves. The number of the dislocations accumulated on an interface is reduced to a half. The division weakens mutual repulsion among converged dislocations.
The voluminous defect accumulating region (H) is a region having a definite thickness. Inner cores (S) can also accommodate dislocations. The dislocation density per unit volume is reduced by additional accumulation in the cores (S). Lower dislocation density in the voluminous defect accumulating regions (H) prevents dislocations from relaxing and escaping.
The conventional ELO method relies upon C-plane growth which maintains a smooth C-plane surface without facets. The dislocations once assembled into the planar defects (bisecting planes) are not constricted and begin to disentangle themselves from the planar defects. Diffusion of the dislocations proceeds during the growth. Dislocations disperse uniformly in the growing GaN crystal. An average dislocation density is about 107 cm−2 in the GaN crystal obtained by the conventional ELO. The GaN crystal of such a 107 cm−2 high dislocation density is entirely useless for a substrate for making InGaN laser diodes.
This invention succeeds in avoiding burying of facet slopes, in maintaining the facet slopes by forming voluminous defect accumulating regions (H) of a definite thickness, and in captivating dislocations in the voluminous defect accumulating regions (H).
This is a feature of the present invention. What enables the regions (H) to encapsulate dislocations is either polycrystalline voluminous defect accumulating regions (H) or single crystalline voluminous defect accumulating regions (H) having shallow facets on the top.
The defect accumulating regions (H) should have a definite width for permanently arresting dislocations. The “definite width” is signified by a word “voluminous”. Thus, the accumulating regions are called “voluminous” defect accumulating regions (H). The gist of the present invention is to decrease dislocations by growing GaN with voluminous defect accumulating regions (H). The width of the voluminous defect accumulating regions (H) is 1 μm to 200 μm.
The voluminous defect accumulating regions (H) and the low dislocation single crystal regions (Z) occur, satisfying a complementary relation. Controlling positions and sizes of voluminous defect accumulating regions (H) occurring determines positions and sizes of the low dislocation single crystal regions (Z). The positions and sizes of voluminous defect accumulating regions (H) can be predetermined by implanting mask as a seed of growing voluminous defect accumulating regions (H) at an early stage of growth. The seed makes a voluminous defect accumulating region (H) thereupon. A set of facets having slopes is made in the neighborhood of the voluminous defect accumulating regions (H). The facets induce formation of low dislocation single crystal regions (Z) following the facets. Thus, implantation of the seed mask can control the sizes and positions of the low dislocation single crystal regions (Z) via formation of voluminous defect accumulating regions (H).
Motivation of making facet valleys leading voluminous defect accumulating regions (H) depends upon the kinds of the voluminous defect accumulating regions (H). A common motivation is the stripe mask which produces cavities upon the stripes by delaying growth. Growing speed on the mask stripes is lower than the speed on the undersubstrate. The delay of forming surfaces is a reason of making cavities upon the stripes. The cavities stabilize forming and maintaining facet valleys following the stripes.
The voluminous defect accumulating regions (H) has a tendency of inviting occurrence of milder inclining facets thereupon. The milder (shallower) facets form stable valleys made of facets (FIG. 5(b)).
Positions of the valleys are determined. The state having valleys of facets is stable. The valleys are not buried but maintained. Controlling positions of facets is realized by this process. Therefore, positions of low dislocation single crystal regions (Z) and defect accumulating regions (H) are determined and controllable. The low dislocation single crystal regions (Z) and the defect accumulating regions (H) can be regularly arranged. This is one of important points in this invention.
The voluminous defect accumulating regions (H) appear in various versions. Polycrystalline or single crystalline voluminous defect accumulating regions (H) originate from the mask. Polycrystalline voluminous defect accumulating regions (H) discern themselves from the surrounding portions by the difference of a crystalline structure. Single crystal voluminous defect accumulating regions (H) can discriminate themselves from the surrounding portions by existence of interfaces. For example, a single crystal voluminous defect accumulating region (H) is encapsulated by interfaces of planar defects.
The planar defect interface is induced by milder (shallower) sloped facets appearing at an early stage of growth on the top, and the shallow facets make the planar defect interface as interface between two kinds of facets. Cooperation of two different slope facets gather dislocations into the interfaces therebetween, which therefore become a dislocation annihilation/accumulation place.
A conspicuous, unexpected feature is frequently appearing polarity-inversion of voluminous defect accumulating regions (H). The polarity (direction of c-axis) of the voluminous defect accumulating regions (H) is different by 180 degrees from the c-axis of the other low dislocation single crystal regions (Z) and C-plane growth regions (Y). In the inversion case, clear grain boundaries happen at the interfaces between the voluminous defect accumulating regions (H) and the low dislocation single crystal regions (Z). The interface grain boundaries play an active role of accumulating the dislocations swept and gathered by the growing facets. In particular in the case of the polarity-inversion occurring in the voluminous defect accumulating regions (H), controlling of the facet growth can be successfully achieved. The reason is that the region of the polarity-inversion grows more slowly than other regions, the inventors suppose.
The above is the basic principle basing the present invention.
The present invention allows a GaN crystal to solve three mentioned serious problems; the hazy dispersion of diffusing dislocations, the planar defects occurring at the dislocation confluence, and the difficulty of controlling positions of the dislocation confluence. The present invention grows a rack-roof GaN crystal having parallel valleys and hills as shown in FIG. 7 and makes a flat, smooth GaN substrate of low dislocation density as shown in FIG. 8. by mechanical processing the rack-roof GaN crystal.
In FIG. 7, a GaN crystal 22 grown on an undersubstrate 21 has a rack-shaped roof of repetitions of parallel hills and valleys which are steep facets. A voluminous defect accumulating region (H) accompanies a valley of the rack-roof in the vertical direction. Slopes forming the hills and valleys are facets 26. What accompanies the facets 26 in the vertical direction are the low dislocation single crystal regions (Z). FIG. 7 shows a GaN crystal having sharp ridges on the hills without flat C-plane growth regions (Y). In this case, the part held between neighboring voluminous defect accumulating regions (H) is a uniform low dislocation single crystal region (Z). The pitch p, the widths z and h satisfy an equation p=z+h. Otherwise in the case of a GaN including C-plane growth regions (Y), the pitch p, the widths z, y and h satisfy another equation p=2z+y+h. The relation between the height of the hill and the pitch p is described later. FIG. 8 demonstrates a rectangle wafer made by eliminating the undersubstrate from the as-grown GaN substrate, grinding the rack-roof on the top surface and polishing both surfaces of the ground wafer. The GaN wafer has a HZYZHZYZH . . . structure having regularly, periodically aligning voluminous defect accumulating regions (H), low dislocation single crystal regions (Z) and C-plane growth regions (Y). The shape of the C-plane growth regions (Y) depends upon the growth condition. Sometimes the C-plane growth regions (Y) meander with a fluctuating width.